Guy Kewney

Nasty incident in the lab the other day. We had an 'expert' in. My experience of experts was uniformly wonderful (no, really!) until about 20 years ago, when my innocence was stripped away from me by the British Standards Institute.

The lads from the BSI pointed out that our benchmark was a set of Basic programs. In other words, we were testing not the personal computer, but the built-in Basic interpreter (20 years ago, they all had one).

They went away and designed a new benchmark that tested the memory access speed of the machine. It used 'machine code' to read the memory chips. The machine code, it turned out, was done by PEEK and POKE commands. In Basic.

This time, our expert was setting up a test of wireless local area network (Lan) equipment. I suspect his tests will turn out to be useful, because they're designed to spot compatibility failures, but in the end he got so cross with us that he decided not to let us print this data.

Why did he get so cross? He mentioned, in passing, that he had set up a wireless link that was 'only' running at 10Mbps. He suggested that it should, obviously, have a throughput of 54Mbps, because it was an 802.11g link.

It turns out that experience of wired networking, however great, is not a good guide to how wireless runs. It's hard enough to test wired Ethernet Lans, but when it comes to wireless Lans, it's even worse.

The problem is that wireless Ethernet is a standard based on 802.11b, which dates back to a time when the fastest cards ran at 1Mbps. I won't bore you with all the tedious details, but the result is that there is a preamble built into every data frame, which has to be transmitted at 1Mbps.

At 1Mbps, it's a very small part of the transmission. At 11Mbps, the preamble takes up nearly half the transmission. Connect two Windows PCs over a 100Mbps Lan and send a long file (say, 100MB), and you'll discover you are dealing with an apparent transmission speed of less than half that.

Make it a shorter file, and it will appear to take far longer, strolling past the tape at an apparent 15Mbps. Ask the system to move 20 short files, writing all the data to disk in between files, and you'll be down into single figures.

Now move to wireless. Try it on a 11Mbps (nominal) wireless 802.11b link, and you will be astonished to find that you are getting about 3Mbps across the network, even if you're the only user.

The newest Broadcom chips can do rather better, using streamlined protocols to talk to each other, and can take 802.11g to 30Mbps, or even maximum throughput for a sole user.

Even so, with people advertising these things as 100Mbps technology, there are going to be a lot of disappointed users.

And some of these are experts. I'm not saying a general purpose IT sysadmin type should be an expert on the finer details of wireless Lan - indeed, quite the opposite.

But there are people scratching their heads about 'faulty' wireless cards who are not using duff gear - they're simply making false assumptions about 'real world' throughput, based on their 100Mbps wired Ethernet experience.

It's not that the makers are lying. It's just that they all want to be measured by the same tape; and the tape has been stretched by over-use. Oh, and don't believe what they tell you about battery life, either.

Like this story? Spread the news by clicking below:

 del.icio.us     Digg this     reddit!